CDS&E: Harnessing Self-Organizing Maps for the Discovery of Star Formation in Molecular Clouds

像我们的太阳这样的恒星诞生于巨大的尘埃和气体云中，这些尘埃和气体同时形成数千颗恒星。天文学家观察这些气体云来研究年轻恒星，了解恒星和行星是如何形成的。这些云很复杂，观测结果很难解释。这个项目使用机器学习来解决天文学中的一个重要谜团：为什么恒星的总质量变化很大？恒星的质量在太阳质量的十分之一到太阳质量的100倍之间。研究人员将使用恒星形成的计算机模型来测试他们的方法，并将其应用于最近对恒星形成区域的观测。该团队将为公众开发演示文稿，并在德克萨斯州天文学本科生研究经验不足的学生（TAURUS）暑期课程中培训本科生机器学习方法。研究人员计划将联合收割机最先进的数值模拟、无监督机器学习和分子线观测相结合，研究分子云中参与星星形成最早阶段的气体。 该团队将应用一种被称为自组织映射（SOM）的无监督神经网络算法来识别由恒星形成结构和形成它们的气体组成的星团。由于分子云内的气体结构复杂，没有明确的边界，这就需要新的SOM聚类推断技术。 将产生分子线“合成观测”，并与观测数据进行比较，以帮助解释和提取物理性质。这项工作将在天文学、统计学和计算机科学之间建立桥梁，从而推动数据分割和聚类的新方法，这将对天文学界有更广泛的用途。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

Stars like our Sun are born in large clouds of dust and gas that make thousands of stars at once. Astronomers observe these gas clouds to study young stars and learn how stars and planets form. These clouds are complex, and the observations are difficult to interpret. This project uses machine learning to address an important mystery in astronomy: Why do stars have a great variety of total mass? Stars are found to have a mass between a tenth the mass of our Sun and up to 100 times the mass of the Sun. The Investigators will use computer models of forming stars to test their method and will apply it to recent observations of star-forming regions. The team will develop presentations for the public and train undergraduates in the Texas Astronomy Undergraduate Research experience for Under-represented Students (TAURUS) summer program in machine learning methods. The investigators plan to combine state-of-the-art numerical simulations, unsupervised machine learning, and molecular line observations to study the gas engaged in the earliest stages of star formation within molecular clouds. The team will apply an unsupervised neural network algorithm, known as a Self-Organizing Map (SOM), to identify clusters comprising star-forming structures and the gas that forms them in observational data. As gas structures within molecular clouds are complex and do not have well-defined boundaries, this necessitates new techniques for SOM cluster inference. Molecular line “synthetic observations” will be produced and compared to the observational data to aid in interpretation and exaction of physical properties. This work will build bridges between astronomy, statistics and computer science and thereby advance new approaches for data segmentation and clustering that will be of broader use to the astronomy community.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.